The cultural shift we need, and the toolset to accompany it, is a shift to application-oriented validation. Instead of an absolute, universal 'yes' or 'no' for some RDF document, we need a more nuanced approach. An RDF document might be well-suited for use in a photo metadata application, but missing some data that is needed for an addressbook.
I think Dan is onto something really interesting here. He's contrasting the XML view of a document suitable for a single purpose with the RDF view of documents suitable for many purposes whether alone or in combination. The traditional way has been for each application to define their own format forcing the author to duplicate information across multiple incompatible formats. This isn't the RDF way, whose data model allows documents to contain data suitable for multiple applications without duplication. Dan's suggesting that rather than checking for conformance to a specific schema, applications could query the RDF file presented to determine whethet its fit for purpose. He points to Sparql as the tool for the job, perhaps with an online catalogue.
Rather than a catalogue, why not a simple RDF vocabulary that allows people to associate queries with classes of applications?